Optimal intra track-surface adjustments

[DeltaLover]

I am currently revisiting the implementation of my speed figures, focusing on a topic that I find very interesting and sufficiently complex both from the data collection process scope of view and the optimal algorithm to use as well.

The problem lies in the creation an optimal intra track-surface adjustments, which I believe is very poorly addressed in the existing bibliography. Currently I have a few ideas of how to attack the problem although I would like to hear from others about it.

What is your solution?

[raybo]

"Intra" - "within", meaning within the same track. I think the more difficult adjustment is "inter" ("between" or "among"), meaning from track to track.

"Intra" surface adjustments include air temperature, air density, surface density and friction, wind force and direction, path bias changes, etc..

"Inter" surface adjustments include all the above, plus, track configuration differences, turn radii differences, turn slope differences, length of turns differences, length of straights differences, starting gate locations differences, etc..

Good luck with your study/potential solutions.

[DeltaLover]

Originally Posted by raybo

"Intra" - "within", meaning within the same track. I think the more difficult adjustment is "inter" ("between" or "among"), meaning from track to track.

"Intra" surface adjustments include air temperature, air density, surface density and friction, wind force and direction, path bias changes, etc..

"Inter" surface adjustments include all the above, plus, track configuration differences, turn radii differences, turn slope differences, length of turns differences, length of straights differences, starting gate locations differences, etc..

Good luck with your study/potential solutions.

You are right about your definitions. The way I see the problem is to derive adjustments for any kind of track - surface pair though. For example comparing BEL - D to BEL - T or AQU - d to TP - A both fall into the same category.

As far as track configuration differences, turn radii differences, turn slope differences, length of turns differences etc, of course they affect the final times and the raw figures but my approach is purely numeric and do not take any of these factors into account based in the assumption that they are already embedded in them.

From a pure algorithmic scope of view. the solution to the problem ranges from the naive approach that is described in Beyer's books to a some very sophisticated methodologies involving graphs, simplex method etc..

Starting this thread, I was curious about how others are attacking this challenging and pragmatic problem but as it turns out there is no interest in this topic as the PA community seems to be more attracted to threads like What is your favorite type of wager? or even Creed-End of the Rocky movies?

[raybo]

So, by only considering one track, but dirt versus turf, these adjustments will only come into play when horses at that track change surfaces. How often does that happen? Or are you thinking that, in the case of horses who have run both surfaces, at the same track, you are hoping to equalize them in order to not have to ignore past performances due to them being on a different surface than today's race?

[Tom]

No need to get insulting about it.
Maybe most of us here do not see a problem with the ways in use now - TFUS, T-Graph/Rags, Beyer, HTR, J-Capper.

If you thing there is something better, you have proven standards to measure the results, which are all that matter.

[cj]

Don't get your panties in a wad. The thread has been up a day, and it was a very busy day of racing. It would have drawn some interest I'm sure. Now, who knows?

[DeltaLover]

Originally Posted by raybo

So, by only considering one track, but dirt versus turf, these adjustments will only come into play when horses at that track change surfaces. How often does that happen? Or are you thinking that, in the case of horses who have run both surfaces, at the same track, you are hoping to equalize them in order to not have to ignore past performances due to them being on a different surface than today's race?

Exactly these are some of the questions that we need to answer in this exercise.

I general, we should start with a hypothesis that I try to support with empirical data. Note that we have several problems to solve here:

• The creation of the raw figures, something that probably involves the development of a track variant, class pars, distance to distance adjustments etc
• The hypothesis of what might be a good way to compare figures across distances and surface. For example we might translate every winning time to its corresponding figure and then assume that shippers who finish first or second in two different tracks within a specific time span are good candidates for adjustment estimation
• The algorithms to use to optimize the adjustment times. For example, in my database I have over 1,800 possible pairs of track - distance and some of them are connected directly by one or more suitable horses while some of them are connected by a certain path (meaning Track X might not be directly connected to Track Y but still there is a track Z that connects them)

As I said before this is a challenging problem to solve efficient and parametrically and this is why I started this thread..

[DeltaLover]

Originally Posted by cj

Don't get your panties in a wad. The thread has been up a day, and it was a very busy day of racing. It would have drawn some interest I'm sure. Now, who knows?

I know CJ, it certainly was a busy day for most of us!

[DeltaLover]

Originally Posted by Tom

No need to get insulting about it.
Maybe most of us here do not see a problem with the ways in use now - TFUS, T-Graph/Rags, Beyer, HTR, J-Capper.

If you thing there is something better, you have proven standards to measure the results, which are all that matter.

I am not sure that there is something better, I am currently trying to see if it can be discovered

[DeltaLover]

To make the conversation more tangible, you can see a list of all the pairs and their adjustments based on my database and my proprietary raw speed figure and method to select matching horses in the following spreadsheet:

http://www.themindofagambler.com/cro...justments.xlsx

The actual methodology of the raw speed figure creation is not important here as the selection of matching horses is neither. The objective is to optimize the adjustments of each pair in an optimal way to minimize errors. A negative adjustment means that the second track - surface pair needs to be subtracted by this number in order to equate the first. Note that we can have a very large number of connecting paths from track to track and how to handle this is a good part of the challenge..

Note that what i call raw figures are already comparable for the same track and surface, so their rawness is only in comparison to other track - surfaces

[PaceAdvantage]

And to be fair, handicapper's corner isn't the most popular section of the site. There are only 12 people viewing this section right now, while there are 75 viewing General Handicapping Discussion. Want me to move to to General Handicapping discussion?

[DeltaLover]

Originally Posted by PaceAdvantage

And to be fair, handicapper's corner isn't the most popular section of the site. There are only 12 people viewing this section right now, while there are 75 viewing General Handicapping Discussion. Want me to move to to General Handicapping discussion?

Sure, I think it makes perfect sense!

[PaceAdvantage]

No problem...thread moved...

[Cratos]

Originally Posted by DeltaLover

I am currently revisiting the implementation of my speed figures, focusing on a topic that I find very interesting and sufficiently complex both from the data collection process scope of view and the optimal algorithm to use as well.

The problem lies in the creation an optimal intra track-surface adjustments, which I believe is very poorly addressed in the existing bibliography. Currently I have a few ideas of how to attack the problem although I would like to hear from others about it.

What is your solution?

The reply in post #2 from the poster, Raybo offers the correct solution if you have the correct empirical data.

However if your hypothesis is based on anecdotal data as the majority if not all of the speed figure methodologies are, the Raybo’s post might be of little help to you.

Therefore you asked a question about “track-surface adjustments” (intra/inner) and my suggestion would be to start with the coefficient of kinetic friction.

For example, if you have calculated the coefficient of kinetic friction for the dirt surface and the coefficient of kinetic friction for the turf surface at a particular racetrack or two different racetracks the next step would to determine the normal force for the horse(s) in question and multiply that normal force by the respective coefficients of kinetic friction.

The difference when divided by the horses’ total mass will give you acceleration in m/seconds.

Convert that into feet/second and divide by the speed of the race of each horse measured in feet/second and you will have the speed loss (so-called track variant) determined by the kinetic friction of both the dirt surface and the turf surface.

Some useful equations:

D (distance) = (1/2) at squared

Velocity =at

Where t = time and a = acceleration

[DeltaLover]

Originally Posted by Cratos

The reply in post #2 from the poster, Raybo offers the correct solution if you have the correct empirical data.

However if your hypothesis is based on anecdotal data as the majority if not all of the speed figure methodologies are, the Raybo’s post might be of little help to you.

Therefore you asked a question about “track-surface adjustments” (intra/inner) and my suggestion would be to start with the coefficient of kinetic friction.

For example, if you have calculated the coefficient of kinetic friction for the dirt surface and the coefficient of kinetic friction for the turf surface at a particular racetrack or two different racetracks the next step would to determine the normal force for the horse(s) in question and multiply that normal force by the respective coefficients of kinetic friction.

The difference when divided by the horses’ total mass will give you acceleration in m/seconds.

Convert that into feet/second and divide by the speed of the race of each horse measured in feet/second and you will have the speed loss (so-called track variant) determined by the kinetic friction of both the dirt surface and the turf surface.

Some useful equations:

D (distance) = (1/2) at squared

Velocity =at

Where t = time and a = acceleration

My approach is completely different and not based on physics but on statistics and algorithms rather, assuming that the chaotic details of surface friction, air resistance, humidity etc are already embedded in the final times . The problem as I defined in my previous posts lies in the selection of the proper data structures and graph algorithms..